Automated Identification of Moderate-to-Severe Traumatic Brain Injury Lesions (AIMS-TBI) MICCAI 2024 Challenge

T1-weighted (T1-w) anatomical magnetic resonance imaging (MRI) enable us to identify injuries (e.g., extent, location, type of lesion, size etc.) within the brains of individuals with moderate-to-severe traumatic brain injury (ms-TBI). Lesion segmentation is a key step prior to running advanced neuroimaging analyses (such as connectomics, tractography); however, to date, no automated lesion segmentation tools have been developed for T1-w images of patients with ms-TBI. To find a solution to this, we established the Automated Identification of Moderate-to-Severe TBI (AIMS-TBI) challenge for MICCAI 2024. For this challenge, the Enhancing NeuroImaging Genetics through Meta-Analysis (ENIGMA) Brain Injury working group collated 1100 T1-w scans from individuals with ms-TBI across 12 sites. Data from 764 individuals underwent manual lesion segmentation using a team of 12 manual raters. For these analyses, 388 images were used in the training dataset, 101 for validation, and 275 for the final test set. During the validation phase, 12 submissions were received, and five teams submitted the final test set. To identify the algorithm that generated the most accurate lesion segmentation from the teams, four metrics were computed, including Dice score, absolute volume difference, absolute lesion count difference, and the lesion-wise F1 score. Scores were calculated for each image, and then averaged, with teams ranked according to best cumulative performance across all metrics.